3D Shape Completion Using Patch Retrieval and Deformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems face challenges in accurately and flexibly generating complete three-dimensional digital shapes due to limitations in inferring correlations between missing and observed surfaces, difficulty in synthesizing geometric styles with large variations, and cubic scaling of voxel counts, leading to inaccurate and unrealistic digital shapes.
Innovation Solution
A data-driven shape completion system using machine learning models for patch retrieval and deformation, which includes a two-stage process of learning to retrieve candidate patches and deforming them to seamlessly blend into the complete shape, leveraging neural networks to guide the process and preserve local geometric features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional voxel-based methods are used to complete 3D shapes, then the approach is simple to implement, but the manufacturing precision and realism of the completed shapes deteriorate due to cubic scaling of voxel counts and inability to capture fine geometric details
Solution Approach 1:
The patent segments the 3D shape completion problem into two distinct stages: (1) coarse shape completion using voxel-based methods to capture overall structure, and (2) fine detail completion using patch-based neural network methods to restore geometric details. This segmentation allows each method to operate at its optimal resolution level, avoiding the cubic scaling problem while preserving fine details.
Solution Approach 2:
The patent transitions from working solely in voxel space (discrete 3D grid) to incorporating continuous patch representations. By retrieving and deforming patches in continuous 3D space and blending them into the voxel-based coarse shape, the system achieves fine geometric detail without being constrained by voxel resolution, effectively adding a dimensional layer of detail representation.
2Manufacturing precision
If neural network-based patch retrieval and deformation is used, then the accuracy and realism of completed shapes is improved, but the device complexity and computational requirements increase
Solution Approach 1:
The patent performs preliminary coarse shape completion using simple voxel-based methods before applying the complex neural network patch retrieval and deformation processes. This preliminary action establishes a structural framework that guides subsequent detailed work, reducing the search space and computational burden of the neural network operations.
Solution Approach 2:
The patent applies complex neural network-based patch retrieval and deformation only to specific regions where fine geometric details are needed, rather than processing the entire 3D shape uniformly. By identifying and focusing computational resources on regions requiring high precision, the system achieves accurate shape completion while minimizing overall computational complexity.
3Device complexity
If conventional systems attempt to infer correlations between missing and observed surfaces, then the process can be simplified, but the reliability and accuracy of the inferred geometry deteriorates due to inability to capture complex geometric styles
Solution Approach 1:
The patent copies existing geometric patterns and styles from observed surfaces of the input shape and applies them to missing regions through patch retrieval. By copying actual geometric features rather than inferring them through simple correlation, the system preserves the authentic geometric style and structural correlations present in the original object, significantly improving reliability of the completed shape.
Data Source
AI summary
Methods, systems, and non-transitory computer readable storage media are disclosed that utilizes machine learning models for patch retrieval and deformation in completing three-dimensional digital shapes. In particular, in one or more implementations the disclosed systems utilize a machine learning model to predict a coarse completion shape from an incomplete 3D digital shape. The disclosed systems sample coarse 3D patches from the coarse 3D digital shape and learn a shape distance function to retrieve detailed 3D shape patches in the input shape. Moreover, the disclosed systems learn a deformation for each retrieved patch and blending weights to integrate the retrieved patches into a continuous surface.